基于谐波小波包和BSA优化LS-SVM的铣刀磨损状态识别研究  

State Recognition of Milling Tool Wears Based on Harmonic Wavelet Packet and BSA Optimization LS-SVM

作　　者：董彩云[1] 张超勇[1] 孟磊磊[1] 肖鹏飞[1] 罗敏[2] 林文文[3] 

机构地区：[1]华中科技大学机械学院数字制造装备与技术国家重点实验室,武汉430074 [2]湖北汽车工业学院电气与信息工程学院,十堰442002 [3]宁波大学机械工程与力学学院,宁波315211

出　　处：《中国机械工程》2017年第17期2080-2089,2108,共11页China Mechanical Engineering

基　　金：国家自然科学基金资助项目(51575211;51421062);国家自然科学基金国际(地区)合作与交流项目(51561125002);湖北省自然科学基金资助项目(2014CFB348);高等学校学科创新引智计划资助项目(B16019)

摘　　要：针对铣削刀具磨损状态识别问题,提出谐波小波包和最小二乘支持向量机(LS-SVM)的状态识别方法。为克服传统小波包分解的频带交叠问题,采用谐波小波包提取不同磨损状态下铣削力信号的各频段信号能量,归一化处理后,输入LS-SVM多类分类器,实现铣削刀具磨损状态的识别。针对LS-SVM的惩罚因子和核参数对模型识别精度影响较大的问题,提出回溯搜索算法(BSA)进行自动参数寻优。实验结果表明,谐波小波包比小波包在刀具磨损状态特征提取时具有更好的识别效果。与粒子群算法进行比较,证明BSA优化LS-SVM具有更高的识别精度。Aiming at the problems of milling tool wear state recognitions,a state recognition method was proposed based on harmonic wavelet packet and LS-SVM.To overcome the band overlapping problems in traditional wavelet packet decompositions,the milling force signal energies of each bands were extracted in different wear states by harmonic wavelet packet,which were brought in multi-class LS-SVM classifier after normalizing,then the classification recognition of different cutting tool states was achieved.BSA was proposed to search the optimal values of the kernel functional parameters and error penalty factors which affected the precision of the LS-SVM significantly.Experimental results show that harmonic wavelet packet is more effective and feasible than wavelet packet,and the proposed milling tool wear recognition method has higher accuracy.

关 键 词：刀具磨损 谐波小波包 回溯搜索算法 最小二乘支持向量机 

分 类 号：TH164[机械工程—机械制造及自动化]